Over the course of the past year, most AI developers have learned how to configure their coding environment, optimizing for the 3rd party packages/integrations, specialized processes, even personal preferences and shortcuts by adopting generally IDE/language/framework agnostic primitives. MCP was a catalyst to jumpstart the context engineering revolution. In its infancy, even the experts had their fair share of complaints and points of confusion. So, it's comical that the next major innovations centered around simple text files like AGENT.md, CLAUDE.md, MEMORY.md, and SKILLS.md. Exact configurations, formatting, file structure, etc. may vary slightly between providers, skills marketplaces, etc. but overall, with the exception of bundled scripts for skills, most .md files are universally compatible between different systems; no coding required to create/modify the .md primitives.

Now, it's about at this point where things get interesting. Honestly, when I decided to leave my corporate job to AI all day, I was surprised and caught off guard to have been faced with a seriously consequential decision right away. From this point on, my personality and everything I am in this world, would all be defined by my choice to MCP or not MCP. You know what they say, "you're only as good as your hobbling along agent..." something like that. You're either an MCP person or anti-MCP person. That's it, two options, no room in the middle, you're either in or you're out...

Example Chat Message:

"I just watched a video and the YouTube fella already deleted his MCP servers live on his video and he do it, so must be right for me too...do what he do...DELETE! "

"Next, create markdown files to replace all the MCP servers and tools we just deleted. /ralph-loop" Muwahahaha GIVE ME SKILLS!

Now, even if you've never heard of MCP, SKILLS.md, etc. but have experience designing, building, etc. really any type of project based work or creative space, you already know, these things never work that way, at least not for me. See Appendix A for an example of my game-changing decision analysis process.

Personally, I'm not going to delete any of the primitives or MCP servers, but when I hear the anti-MCP crowd voice frustrations, I imagine an interaction like this:

Here's what I keep coming back to after every MSP conversation about AI: you've already done this before.

Cloud and Hybrid changed the architecture of IT Infrastructure. It expanded the edge. It blurred the lines of what MSPs take ownership of for their clients. And the MSPs who moved first — who built competency before the market demanded it — are the ones who thrived.

AI is the same pattern. Different technology, same playbook.

There's no shortage of AI hype in the MSP space. Every vendor has an AI story. Every conference has an AI track. But when you sit down with MSPs who are actually using this stuff in production — not demoing it, not piloting it, using it — the playbook looks very different from the marketing slides.

Here's what's working, what's not, and where the highest-leverage opportunities are right now.

Just like the evolution of Cloud and Hybrid many years ago, AI is changing the architecture of IT Infrastructure, expanding the edge, and blurring the lines of what MSPs take ownership of for their clients. While AI technology still feels abstract to many, those responsible must apply the same controls across the stack — in terms of security and data governance — that we always have.

It pains me when I hear about AI training programs that focus on educating employees how to avoid uploading sensitive material. While awareness training shows good intention, it's insufficient without architectural changes. Business Email Compromise hasn't been eliminated — it's evolving. We've spent so much time, money, and energy establishing security standards. So why are we treating AI differently?

While this message may sound like harsh criticism, I do live in the real world. I understand these are not easy problems to address. But the regulatory landscape isn't waiting for us to figure it out.

I recently had the privilege of joining and speaking with roughly 100 leaders of Managed Services Providers — direct competitors, all gathered in the same room to discuss the current state, desired future state, and unknowns of leveraging Artificial Intelligence in their businesses. It's inspiring to see competitors supporting each other in this journey.

The AI conversation in the MSP space is no longer "if." It's how, where, and how fast it pays back. The FOMO is real, but so are the results.

Exciting news from Anthropic! Claude Code now supports plugins—customizable collections that bundle slash commands, agents, MCP servers, and hooks that can be installed with a single command. Even better, this announcement features an open source project I contributed over 100 agent templates to!

N-able recently announced Cat-MIP, a groundbreaking initiative to standardize terminology for AI automation across the cybersecurity and IT management landscape. This is particularly relevant for anyone working with AI agents and Model Context Protocol (MCP) servers.

The LangChain Interrupt 2025 conference recently brought together leading minds in Artificial Intelligence to explore the fast-evolving world of AI agents. A standout theme was the rise of the "agent engineer"—a new kind of builder skilled in prompting, product development, software engineering, and machine learning, all crucial for creating the next generation of intelligent systems.

While AI advances at rapid rate, the basics remain the same.